A Data Driven Approach For Multimodal Process Monitoring Using Bayesian Analytics
Dr. Hariprasad K
9th June 2017, Friday, 12:00-1:00 pm, ChE Seminar hall
Process monitoring is imperative to maintain safe and reliable operation of chemical process by detection and diagnosis of root causes of anomalies or faults in process operations. It is one of the research areas where process control community has widely used data driven approaches due to the difficulties in obtaining first principles based models. The initial research directions in this area focused on process monitoring based on univariate control charts. Later, large dimensionality and correlations among the variables of the chemical process data demanded researchers to look for multi-variate approaches. This has led to deployment of models such as Principal Component Analyzers (PCA) that project high dimensional process data into a low-dimensional space, enabling lower dimensional control charts while preserving correlations. However, predominant latent variable models implicitly assume Gaussian distribution for process variables making it suitable only for processes that exhibit unimodal characteristics. In reality, processes have multimodal characteristics arising from multi-mode operations, process nonlinearities and non-stationarities, making process monitoring a challenging problem. In this talk, some of these challenges associated with multimodal process monitoring and our recent research for addressing these challenges using mixtures of Probabilistic Principal Component Analyzers (PPCA) under Bayesian framework would be discussed. Our modeling approach facilitates in obtaining interpretable models for complex data distributions, thereby making it viable for process monitoring tasks such as fault detection and diagnosis.
ABOUT THE SPEAKER
Dr. Hariprasad K is a postdoctoral fellow with Computer Process Control group, Department of Chemical and Materials Engineering, University of Alberta since February 2015. Prior to that he completed his PhD from IIT Bombay in December 2014. His thesis was awarded “R G Manudhane PhD excellence award” and “Award for excellence in PhD thesis” from IIT Bombay. His research interests include model predictive control, robust system identification and Bayesian analytics applied to process modeling and monitoring.